Add contrib groupnorm (#3678)

GroupNorm Contrib Operator from Microsoft's Contrib set: https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.GroupNorm

Needed for additional support on optimized Stable diffusion models outlined here: #3538

This operator maps almost all inputs 1:1 with our existing groupNormalization operator and renames a few fields/attributes

src/onnx/parse_groupnorm.cpp

@@ -1,7 +1,7 @@
 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2024 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a copy
  * of this software and associated documentation files (the "Software"), to deal
@@ -25,43 +25,96 @@
 #include <migraphx/ranges.hpp>
 #include <migraphx/make_op.hpp>
 #include <migraphx/instruction.hpp>
+#include <migraphx/permutation.hpp>
 
 namespace migraphx {
 inline namespace MIGRAPHX_INLINE_NS {
 namespace onnx {
 
+static instruction_ref
+apply_channels_last_perm(const onnx_parser::node_info& info, instruction_ref ins, bool invert)
+{
+    std::vector<int64_t> perm(ins->get_shape().ndim());
+    std::iota(perm.begin() + 1, perm.end() - 1, 2);
+    perm.back() = 1;
+    return info.add_instruction(
+        make_op("transpose", {{"permutation", invert ? invert_permutation(perm) : perm}}), ins);
+}
+
 struct parse_groupnorm : op_parser<parse_groupnorm>
 {
-    std::vector<op_desc> operators() const { return {{"GroupNormalization"}}; }
+    std::vector<op_desc> operators() const
+    {
+        return {{"GroupNormalization", "GroupNorm"}, {"GroupNorm", "Contrib_GroupNorm"}};
+    }
 
-    instruction_ref parse(const op_desc& /*opd*/,
+    instruction_ref parse(const op_desc& opd,
                           const onnx_parser& parser,
                           const onnx_parser::node_info& info,
                           std::vector<instruction_ref> args) const
     {
+        bool is_contrib = (opd.op_name == ("Contrib_GroupNorm"));
+
         float epsilon = 1e-5f;
         if(contains(info.attributes, "epsilon"))
         {
             epsilon = parser.parse_value(info.attributes.at("epsilon")).at<float>();
         }
         size_t num_groups;
-        if(contains(info.attributes, "num_groups"))
+        if(contains(info.attributes, "num_groups") or contains(info.attributes, "groups"))
         {
-            num_groups = parser.parse_value(info.attributes.at("num_groups")).at<size_t>();
+            if(is_contrib)
+            {
+                num_groups =
+                    std::abs(parser.parse_value(info.attributes.at("groups")).at<int64_t>());
+            }
+            else
+            {
+                num_groups =
+                    std::abs(parser.parse_value(info.attributes.at("num_groups")).at<int64_t>());
+            }
         }
         else
         {
             MIGRAPHX_THROW("PARSE_GROUPNORM: num_groups must be available");
         }
 
+        bool is_channels_last = false;
+        if(is_contrib)
+        { // default state for GroupNorm Contrib op
+            is_channels_last = true;
+            if(contains(info.attributes, "channels_last"))
+            {
+                is_channels_last =
+                    (1 == parser.parse_value(info.attributes.at("channels_last")).at<size_t>());
+            }
+        }
+
+        bool silu_activation = false;
+        if(contains(info.attributes, "activation") and is_contrib)
+        {
+            silu_activation =
+                (1 == parser.parse_value(info.attributes.at("activation")).at<size_t>());
+        }
+        else if(is_contrib)
+        {
+            MIGRAPHX_THROW("PARSE_GROUPNORM: activation must be available");
+        }
+
         if(args.size() != 3)
         {
             MIGRAPHX_THROW("PARSE_GROUPNORM: invalid input count");
         }
 
-        auto x     = args.at(0);
-        auto scale = args.at(1);
-        auto bias  = args.at(2);
+        // Adjust chanels from channels_last-> NCHW if last channel is set for contrib op
+        auto x = args.at(0);
+        if(is_channels_last and is_contrib)
+        {
+            x = apply_channels_last_perm(info, x, true);
+        }
+
+        auto scale = args.at(1); // gamma in the GroupNorm contrib case
+        auto bias  = args.at(2); // beta in the GroupNorm contrib case
 
         auto x_shape = x->get_shape();
         auto x_dtype = x_shape.type();
@@ -120,7 +173,20 @@ struct parse_groupnorm : op_parser<parse_groupnorm>
             info.add_instruction(make_op("broadcast", {{"axis", 1}, {"out_lens", dims}}), bias);
         auto scaled = info.add_instruction(make_op("mul"), result, scale_bcast);
         auto y      = info.add_instruction(make_op("add"), scaled, bias_bcast);
-        return info.add_instruction(make_op("reshape", {{"dims", x_dims}}), y);
+        auto output = info.add_instruction(make_op("reshape", {{"dims", x_dims}}), y);
+
+        // Convert to NCHW -> channels_last for contrib GroupNorm
+        if(is_channels_last and is_contrib)
+        {
+            output = apply_channels_last_perm(info, output, false);
+        }
+        if(silu_activation)
+        {
+            // SiLU activation is just  out = x * sigmoid(x)
+            auto sigmoid = info.add_instruction(make_op("sigmoid"), output);
+            output       = info.add_instruction(make_op("mul"), output, sigmoid);
+        }
+        return output;
     }
 };
 

test/onnx/gen_onnx.py

@@ -4862,6 +4862,112 @@ def group_norm_invalid_bias_shape_test():
     return group_norm_test([1, 4, 3, 3], [2], [3], [1, 4, 3, 3], 2)
 
 
+def group_norm_contrib_test(x_dims,
+                            gamma_dims,
+                            beta_dims,
+                            y_dims,
+                            num_groups,
+                            activation,
+                            channels_last,
+                            eps_value=1e-5,
+                            dtype=TensorProto.FLOAT):
+    x = helper.make_tensor_value_info('x', dtype, x_dims)
+    gamma = helper.make_tensor_value_info('gamma', dtype, gamma_dims)
+    beta = helper.make_tensor_value_info('beta', dtype, beta_dims)
+    y = helper.make_tensor_value_info('y', dtype, y_dims)
+
+    node = onnx.helper.make_node('GroupNorm',
+                                 inputs=['x', 'gamma', 'beta'],
+                                 outputs=['y'],
+                                 activation=activation,
+                                 channels_last=channels_last,
+                                 groups=num_groups,
+                                 epsilon=eps_value)
+
+    return ([node], [x, gamma, beta], [y])
+
+
+@onnx_test()
+def group_norm_contrib_3d_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2], 2, 0, 0)
+
+
+@onnx_test()
+def group_norm_contrib_3d_channel_last_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2], 2, 0, 1)
+
+
+@onnx_test()
+def group_norm_contrib_3d_channel_last_half_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2],
+                                   2,
+                                   0,
+                                   1,
+                                   dtype=TensorProto.FLOAT16)
+
+
+@onnx_test()
+def group_norm_contrib_3d_channel_last_bf16_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2],
+                                   2,
+                                   0,
+                                   1,
+                                   dtype=TensorProto.BFLOAT16)
+
+
+@onnx_test()
+def group_norm_contrib_silu_3d_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2], 2, 1, 0)
+
+
+@onnx_test()
+def group_norm_contrib_channels_last_3d_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2], 2, 0, 1)
+
+
+@onnx_test()
+def group_norm_contrib_channels_last_4d_test():
+    return group_norm_contrib_test([1, 3, 3, 4], [2], [2], [1, 3, 3, 4], 2, 0,
+                                   1)
+
+
+@onnx_test()
+def group_norm_contrib_channels_last_and_silu_3d_test():
+    return group_norm_contrib_test([1, 4, 2], [2], [2], [1, 4, 2], 2, 1, 1)
+
+
+@onnx_test()
+def group_norm_contrib_no_activation_attr_test():
+    x = helper.make_tensor_value_info('x', TensorProto.FLOAT, [1, 4, 2])
+    gamma = helper.make_tensor_value_info('gamma', TensorProto.FLOAT, [2])
+    beta = helper.make_tensor_value_info('beta', TensorProto.FLOAT, [2])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [1, 4, 2])
+
+    node = onnx.helper.make_node('GroupNorm',
+                                 inputs=['x', 'gamma', 'Beta'],
+                                 outputs=['y'],
+                                 channels_last=0,
+                                 groups=2)
+
+    return ([node], [x, gamma, beta], [y])
+
+
+@onnx_test()
+def group_norm_contrib_no_num_groups_attr_test():
+    x = helper.make_tensor_value_info('x', TensorProto.FLOAT, [1, 4, 2])
+    gamma = helper.make_tensor_value_info('gamma', TensorProto.FLOAT, [2])
+    beta = helper.make_tensor_value_info('beta', TensorProto.FLOAT, [2])
+    y = helper.make_tensor_value_info('y', TensorProto.FLOAT, [1, 4, 2])
+
+    node = onnx.helper.make_node('GroupNorm',
+                                 inputs=['x', 'gamma', 'Beta'],
+                                 outputs=['y'],
+                                 activation=0,
+                                 channels_last=0)
+
+    return ([node], [x, gamma, beta], [y])
+
+
 @onnx_test()
 def group_query_attention_test():
     qkv = helper.make_tensor_value_info('qkv', TensorProto.FLOAT16,

test/onnx/group_norm_contrib_channels_last_and_silu_3d_test.onnx

@@ -0,0 +1,29 @@
+
+1group_norm_contrib_channels_last_and_silu_3d_test:�

+i
+
x
+gamma
+beta
y"	GroupNorm*
+
+activation
�
*
+channels_last
�
*
+epsilon��'7�

*
+groups�
1group_norm_contrib_channels_last_and_silu_3d_testZ
+
x
+

+

+
+Z
+gamma
+
+

+Z
+beta
+
+

+b
+
y
+

+

+
+B

test/onnx/include/onnx_test_utils.hpp

@@ -120,14 +120,16 @@ make_group_norm(const std::vector<int64_t>& input_dims,
                 const std::vector<int64_t>& reshape_dims,
                 const std::vector<int64_t>& reduce_axes,
                 const float eps_value               = 1e-5f,
-                const migraphx::shape::type_t dtype = migraphx::shape::float_type)
+                const migraphx::shape::type_t dtype = migraphx::shape::float_type,
+                const std::string& param1_name      = "scale",
+                const std::string& param2_name      = "bias")
 {
     migraphx::program p;
     auto* mm = p.get_main_module();
 
     auto x     = mm->add_parameter("x", {dtype, input_dims});
-    auto scale = mm->add_parameter("scale", {dtype, scale_dims});
-    auto bias  = mm->add_parameter("bias", {dtype, bias_dims});
+    auto scale = mm->add_parameter(param1_name, {dtype, scale_dims});
+    auto bias  = mm->add_parameter(param2_name, {dtype, bias_dims});
 
     auto eps = mm->add_literal(migraphx::literal{dtype, {eps_value}});
 

test/onnx/include/onnx_verify_utils.hpp

@@ -1,7 +1,7 @@
 /*
  * The MIT License (MIT)
  *
- * Copyright (c) 2015-2023 Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a copy
  * of this software and associated documentation files (the "Software"), to deal
@@ -33,7 +33,9 @@ template <typename T = float>
 std::vector<T> norm_test(const std::vector<size_t>& x_dims,
                          std::vector<T>& scale,
                          std::vector<T>& bias,
-                         migraphx::program p)
+                         migraphx::program p,
+                         const std::string& scale_str = std::string{"scale"},
+                         const std::string& bias_str  = std::string{"bias"})
 {
     p.compile(migraphx::make_target("ref"));
 
@@ -46,8 +48,8 @@ std::vector<T> norm_test(const std::vector<size_t>& x_dims,
 
     migraphx::parameter_map pp;
     pp["x"]     = migraphx::argument(s_x, x.data());
-    pp["scale"] = migraphx::argument(s_s, scale.data());
-    pp["bias"]  = migraphx::argument(s_b, bias.data());
+    pp[scale_str] = migraphx::argument(s_s, scale.data());
+    pp[bias_str]  = migraphx::argument(s_b, bias.data());
 
     auto result = p.eval(pp).back();
 

test/onnx/parse/group_norm_contrib_3d_no_activation_err_test.cpp

@@ -0,0 +1,31 @@
+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include <onnx_test.hpp>
+#include <onnx_test_utils.hpp>
+
+TEST_CASE(group_norm_contrib_no_activation_err_test)
+{
+    EXPECT(test::throws([&] { read_onnx("group_norm_contrib_no_activation_attr_test.onnx"); }));
+}

test/onnx/parse/group_norm_contrib_3d_no_num_groups_err_test.cpp

@@ -0,0 +1,31 @@
+/*
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2025 Advanced Micro Devices, Inc. All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include <onnx_test.hpp>
+#include <onnx_test_utils.hpp>
+
+TEST_CASE(group_norm_contrib_no_num_groups_err_test)
+{
+    EXPECT(test::throws([&] { read_onnx("group_norm_contrib_no_num_groups_attr_test.onnx"); }));
+}